When should Data Binning be used in data processing

Question

In data pre-processing, Data Binning is a technique to convert continuous values of a feature to categorical ones. For example, sometimes, the values of age feature in datasets are replaced with one of intervals such as:

[10,20),
[20,30),
[30,40].

When is the best time to use Data Binning? Does it (always) lead to a better result in a predication system or it may work as a trial and error?

Answer 1

Mostly by trial and error. When a continuous variable is binned, some information is automatically discarded. Many algorithms would like to make a forecast with a continuous input, and many would bin the continuous input themselves. If your continuous variable is noisy, meaning the values for your variable were not recorded very properly, binning would be a good idea to use. Binning could then be used to lessen the noise. Equal width binning and equal frequency binning are two binning procedures. When your continuous variable is poorly distributed, I would avoid equal width binning.